// scalarizing them if the elements of the aggregate are only loaded. Note that
// by default it refuses to scalarize aggregates which would require passing in
// more than three operands to the function, because passing thousands of
-// operands for a large array or structure is unprofitable! This limit is can be
+// operands for a large array or structure is unprofitable! This limit can be
// configured or disabled, however.
//
// Note that this transformation could also be done for arguments that are only
#define DEBUG_TYPE "argpromotion"
#include "llvm/Transforms/IPO.h"
-#include "llvm/Constants.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Module.h"
-#include "llvm/CallGraphSCCPass.h"
-#include "llvm/Instructions.h"
+#include "llvm/ADT/DepthFirstIterator.h"
+#include "llvm/ADT/Statistic.h"
+#include "llvm/ADT/StringExtras.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/Analysis/CallGraph.h"
-#include "llvm/Target/TargetData.h"
-#include "llvm/Support/CallSite.h"
+#include "llvm/CallGraphSCCPass.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/LLVMContext.h"
+#include "llvm/IR/Module.h"
#include "llvm/Support/CFG.h"
+#include "llvm/Support/CallSite.h"
#include "llvm/Support/Debug.h"
-#include "llvm/ADT/DepthFirstIterator.h"
-#include "llvm/ADT/Statistic.h"
-#include "llvm/ADT/StringExtras.h"
-#include "llvm/Support/Compiler.h"
+#include "llvm/Support/raw_ostream.h"
#include <set>
using namespace llvm;
namespace {
/// ArgPromotion - The 'by reference' to 'by value' argument promotion pass.
///
- struct VISIBILITY_HIDDEN ArgPromotion : public CallGraphSCCPass {
+ struct ArgPromotion : public CallGraphSCCPass {
virtual void getAnalysisUsage(AnalysisUsage &AU) const {
AU.addRequired<AliasAnalysis>();
- AU.addRequired<TargetData>();
CallGraphSCCPass::getAnalysisUsage(AU);
}
- virtual bool runOnSCC(const std::vector<CallGraphNode *> &SCC);
+ virtual bool runOnSCC(CallGraphSCC &SCC);
static char ID; // Pass identification, replacement for typeid
- ArgPromotion(unsigned maxElements = 3) : CallGraphSCCPass((intptr_t)&ID),
- maxElements(maxElements) {}
-
+ explicit ArgPromotion(unsigned maxElements = 3)
+ : CallGraphSCCPass(ID), maxElements(maxElements) {
+ initializeArgPromotionPass(*PassRegistry::getPassRegistry());
+ }
+
/// A vector used to hold the indices of a single GEP instruction
typedef std::vector<uint64_t> IndicesVector;
private:
- bool PromoteArguments(CallGraphNode *CGN);
+ CallGraphNode *PromoteArguments(CallGraphNode *CGN);
bool isSafeToPromoteArgument(Argument *Arg, bool isByVal) const;
- Function *DoPromotion(Function *F,
- SmallPtrSet<Argument*, 8> &ArgsToPromote,
- SmallPtrSet<Argument*, 8> &ByValArgsToTransform);
+ CallGraphNode *DoPromotion(Function *F,
+
SmallPtrSet<Argument*, 8> &ArgsToPromote,
+
SmallPtrSet<Argument*, 8> &ByValArgsToTransform);
/// The maximum number of elements to expand, or 0 for unlimited.
unsigned maxElements;
};
}
char ArgPromotion::ID = 0;
-static RegisterPass<ArgPromotion>
-X("argpromotion", "Promote 'by reference' arguments to scalars");
+INITIALIZE_PASS_BEGIN(ArgPromotion, "argpromotion",
+ "Promote 'by reference' arguments to scalars", false, false)
+INITIALIZE_AG_DEPENDENCY(AliasAnalysis)
+INITIALIZE_AG_DEPENDENCY(CallGraph)
+INITIALIZE_PASS_END(ArgPromotion, "argpromotion",
+ "Promote 'by reference' arguments to scalars", false, false)
Pass *llvm::createArgumentPromotionPass(unsigned maxElements) {
return new ArgPromotion(maxElements);
}
-bool ArgPromotion::runOnSCC(const std::vector<CallGraphNode *> &SCC) {
+bool ArgPromotion::runOnSCC(CallGraphSCC &SCC) {
bool Changed = false, LocalChange;
do { // Iterate until we stop promoting from this SCC.
LocalChange = false;
// Attempt to promote arguments from all functions in this SCC.
- for (unsigned i = 0, e = SCC.size(); i != e; ++i)
- LocalChange |= PromoteArguments(SCC[i]);
+ for (CallGraphSCC::iterator I = SCC.begin(), E = SCC.end(); I != E; ++I) {
+ if (CallGraphNode *CGN = PromoteArguments(*I)) {
+ LocalChange = true;
+ SCC.ReplaceNode(*I, CGN);
+ }
+ }
Changed |= LocalChange; // Remember that we changed something.
} while (LocalChange);
-
+
return Changed;
}
/// example, all callers are direct). If safe to promote some arguments, it
/// calls the DoPromotion method.
///
-bool ArgPromotion::PromoteArguments(CallGraphNode *CGN) {
+CallGraphNode *ArgPromotion::PromoteArguments(CallGraphNode *CGN) {
Function *F = CGN->getFunction();
// Make sure that it is local to this module.
- if (!F || !F->hasInternalLinkage()) return false;
+ if (!F || !F->hasLocalLinkage()) return 0;
// First check: see if there are any pointer arguments! If not, quick exit.
SmallVector<std::pair<Argument*, unsigned>, 16> PointerArgs;
unsigned ArgNo = 0;
for (Function::arg_iterator I = F->arg_begin(), E = F->arg_end();
I != E; ++I, ++ArgNo)
- if (isa<PointerType>(I->getType()))
+ if (I->getType()->isPointerTy())
PointerArgs.push_back(std::pair<Argument*, unsigned>(I, ArgNo));
- if (PointerArgs.empty()) return false;
+ if (PointerArgs.empty()) return 0;
// Second check: make sure that all callers are direct callers. We can't
- // transform functions that have indirect callers.
+ // transform functions that have indirect callers. Also see if the function
+ // is self-recursive.
+ bool isSelfRecursive = false;
for (Value::use_iterator UI = F->use_begin(), E = F->use_end();
UI != E; ++UI) {
- CallSite CS = CallSite::get(*UI);
- if (!CS.getInstruction()) // "Taking the address" of the function
- return false;
-
- // Ensure that this call site is CALLING the function, not passing it as
- // an argument.
- if (UI.getOperandNo() != 0)
- return false;
+ CallSite CS(*UI);
+ // Must be a direct call.
+ if (CS.getInstruction() == 0 || !CS.isCallee(UI)) return 0;
+
+ if (CS.getInstruction()->getParent()->getParent() == F)
+ isSelfRecursive = true;
}
-
+
// Check to see which arguments are promotable. If an argument is promotable,
// add it to ArgsToPromote.
SmallPtrSet<Argument*, 8> ArgsToPromote;
SmallPtrSet<Argument*, 8> ByValArgsToTransform;
for (unsigned i = 0; i != PointerArgs.size(); ++i) {
- bool isByVal = F->paramHasAttr(PointerArgs[i].second+1, ParamAttr::ByVal);
-
+ bool isByVal=F->getAttributes().
+ hasAttribute(PointerArgs[i].second+1, Attribute::ByVal);
+ Argument *PtrArg = PointerArgs[i].first;
+ Type *AgTy = cast<PointerType>(PtrArg->getType())->getElementType();
+
// If this is a byval argument, and if the aggregate type is small, just
// pass the elements, which is always safe.
- Argument *PtrArg = PointerArgs[i].first;
if (isByVal) {
- const Type *AgTy = cast<PointerType>(PtrArg->getType())->getElementType();
- if (const StructType *STy = dyn_cast<StructType>(AgTy)) {
+ if (StructType *STy = dyn_cast<StructType>(AgTy)) {
if (maxElements > 0 && STy->getNumElements() > maxElements) {
- DOUT << "argpromotion disable promoting argument '"
- << PtrArg->getName() << "' because it would require adding more "
- << "than " << maxElements << " arguments to the function.\n";
- } else {
- // If all the elements are single-value types, we can promote it.
- bool AllSimple = true;
- for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i)
- if (!STy->getElementType(i)->isSingleValueType()) {
- AllSimple = false;
- break;
- }
-
- // Safe to transform, don't even bother trying to "promote" it.
- // Passing the elements as a scalar will allow scalarrepl to hack on
- // the new alloca we introduce.
- if (AllSimple) {
- ByValArgsToTransform.insert(PtrArg);
- continue;
+ DEBUG(dbgs() << "argpromotion disable promoting argument '"
+ << PtrArg->getName() << "' because it would require adding more"
+ << " than " << maxElements << " arguments to the function.\n");
+ continue;
+ }
+
+ // If all the elements are single-value types, we can promote it.
+ bool AllSimple = true;
+ for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
+ if (!STy->getElementType(i)->isSingleValueType()) {
+ AllSimple = false;
+ break;
+ }
+ }
+
+ // Safe to transform, don't even bother trying to "promote" it.
+ // Passing the elements as a scalar will allow scalarrepl to hack on
+ // the new alloca we introduce.
+ if (AllSimple) {
+ ByValArgsToTransform.insert(PtrArg);
+ continue;
+ }
+ }
+ }
+
+ // If the argument is a recursive type and we're in a recursive
+ // function, we could end up infinitely peeling the function argument.
+ if (isSelfRecursive) {
+ if (StructType *STy = dyn_cast<StructType>(AgTy)) {
+ bool RecursiveType = false;
+ for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
+ if (STy->getElementType(i) == PtrArg->getType()) {
+ RecursiveType = true;
+ break;
}
}
+ if (RecursiveType)
+ continue;
}
}
if (isSafeToPromoteArgument(PtrArg, isByVal))
ArgsToPromote.insert(PtrArg);
}
-
- // No promotable pointer arguments.
- if (ArgsToPromote.empty() && ByValArgsToTransform.empty()) return false;
- Function *NewF = DoPromotion(F, ArgsToPromote, ByValArgsToTransform);
-
- // Update the call graph to know that the function has been transformed.
- getAnalysis<CallGraph>().changeFunction(F, NewF);
- return true;
-}
+ // No promotable pointer arguments.
+ if (ArgsToPromote.empty() && ByValArgsToTransform.empty())
+ return 0;
-/// IsAlwaysValidPointer - Return true if the specified pointer is always legal
-/// to load.
-static bool IsAlwaysValidPointer(Value *V) {
- if (isa<AllocaInst>(V) || isa<GlobalVariable>(V)) return true;
- if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(V))
- return IsAlwaysValidPointer(GEP->getOperand(0));
- if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V))
- if (CE->getOpcode() == Instruction::GetElementPtr)
- return IsAlwaysValidPointer(CE->getOperand(0));
-
- return false;
+ return DoPromotion(F, ArgsToPromote, ByValArgsToTransform);
}
-/// AllCalleesPassInValidPointerForArgument - Return true if we can prove that
+/// AllCallersPassInValidPointerForArgument - Return true if we can prove that
/// all callees pass in a valid pointer for the specified function argument.
-static bool AllCalleesPassInValidPointerForArgument(Argument *Arg) {
+static bool AllCallersPassInValidPointerForArgument(Argument *Arg) {
Function *Callee = Arg->getParent();
unsigned ArgNo = std::distance(Callee->arg_begin(),
// have direct callees.
for (Value::use_iterator UI = Callee->use_begin(), E = Callee->use_end();
UI != E; ++UI) {
- CallSite CS = CallSite::get(*UI);
- assert(CS.getInstruction() && "Should only have direct calls!");
+ CallSite CS(*UI);
+ assert(CS && "Should only have direct calls!");
- if (!IsAlwaysValidPointer(CS.getArgument(ArgNo)))
+ if (!CS.getArgument(ArgNo)->isDereferenceablePointer())
return false;
}
return true;
const ArgPromotion::IndicesVector &Longer) {
if (Prefix.size() > Longer.size())
return false;
- for (unsigned i = 0, e = Prefix.size(); i != e; ++i)
- if (Prefix[i] != Longer[i])
- return false;
- return true;
+ return std::equal(Prefix.begin(), Prefix.end(), Longer.begin());
}
return Low != Set.end() && IsPrefix(*Low, Indices);
}
-/// Mark the given indices (ToMark) as safe in the the given set of indices
+/// Mark the given indices (ToMark) as safe in the given set of indices
/// (Safe). Marking safe usually means adding ToMark to Safe. However, if there
/// is already a prefix of Indices in Safe, Indices are implicitely marked safe
/// already. Furthermore, any indices that Indices is itself a prefix of, are
return;
// Increment Low, so we can use it as a "insert before" hint
- ++Low;
+ ++Low;
}
- // Insert
+ // Insert
Low = Safe.insert(Low, ToMark);
++Low;
// If there we're a prefix of longer index list(s), remove those
// This set contains all the sets of indices that we are planning to promote.
// This makes it possible to limit the number of arguments added.
GEPIndicesSet ToPromote;
-
+
// If the pointer is always valid, any load with first index 0 is valid.
- if(isByVal || AllCalleesPassInValidPointerForArgument(Arg))
+ if (isByVal || AllCallersPassInValidPointerForArgument(Arg))
SafeToUnconditionallyLoad.insert(IndicesVector(1, 0));
// First, iterate the entry block and mark loads of (geps of) arguments as
IndicesVector Operands;
for (Value::use_iterator UI = Arg->use_begin(), E = Arg->use_end();
UI != E; ++UI) {
+ User *U = *UI;
Operands.clear();
- if (LoadInst *LI = dyn_cast<LoadInst>(*UI)) {
- if (LI->isVolatile()) return false; // Don't hack volatile loads
+ if (LoadInst *LI = dyn_cast<LoadInst>(U)) {
+ // Don't hack volatile/atomic loads
+ if (!LI->isSimple()) return false;
Loads.push_back(LI);
// Direct loads are equivalent to a GEP with a zero index and then a load.
Operands.push_back(0);
- } else if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(*UI)) {
+ } else if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(U)) {
if (GEP->use_empty()) {
// Dead GEP's cause trouble later. Just remove them if we run into
// them.
getAnalysis<AliasAnalysis>().deleteValue(GEP);
GEP->eraseFromParent();
- // TODO: This runs the above loop over and over again for dead GEPS
+ // TODO: This runs the above loop over and over again for dead GEPs
// Couldn't we just do increment the UI iterator earlier and erase the
// use?
return isSafeToPromoteArgument(Arg, isByVal);
for (Value::use_iterator UI = GEP->use_begin(), E = GEP->use_end();
UI != E; ++UI)
if (LoadInst *LI = dyn_cast<LoadInst>(*UI)) {
- if (LI->isVolatile()) return false; // Don't hack volatile loads
+ // Don't hack volatile/atomic loads
+ if (!LI->isSimple()) return false;
Loads.push_back(LI);
} else {
// Other uses than load?
} else {
return false; // Not a load or a GEP.
}
-
+
// Now, see if it is safe to promote this load / loads of this GEP. Loading
// is safe if Operands, or a prefix of Operands, is marked as safe.
if (!PrefixIn(Operands, SafeToUnconditionallyLoad))
// to do.
if (ToPromote.find(Operands) == ToPromote.end()) {
if (maxElements > 0 && ToPromote.size() == maxElements) {
- DOUT << "argpromotion not promoting argument '"
- << Arg->getName() << "' because it would require adding more "
- << "than " << maxElements << " arguments to the function.\n";
+ DEBUG(dbgs() << "argpromotion not promoting argument '"
+ << Arg->getName() << "' because it would require adding more "
+ << "than " << maxElements << " arguments to the function.\n");
// We limit aggregate promotion to only promoting up to a fixed number
// of elements of the aggregate.
return false;
SmallPtrSet<BasicBlock*, 16> TranspBlocks;
AliasAnalysis &AA = getAnalysis<AliasAnalysis>();
- TargetData &TD = getAnalysis<TargetData>();
for (unsigned i = 0, e = Loads.size(); i != e; ++i) {
// Check to see if the load is invalidated from the start of the block to
LoadInst *Load = Loads[i];
BasicBlock *BB = Load->getParent();
- const PointerType *LoadTy =
- cast<PointerType>(Load->getPointerOperand()->getType());
- unsigned LoadSize = (unsigned)TD.getTypeStoreSize(LoadTy->getElementType());
-
- if (AA.canInstructionRangeModify(BB->front(), *Load, Arg, LoadSize))
+ AliasAnalysis::Location Loc = AA.getLocation(Load);
+ if (AA.canInstructionRangeModify(BB->front(), *Load, Loc))
return false; // Pointer is invalidated!
// Now check every path from the entry block to the load for transparency.
// To do this, we perform a depth first search on the inverse CFG from the
// loading block.
- for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI)
+ for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
+ BasicBlock *P = *PI;
for (idf_ext_iterator<BasicBlock*, SmallPtrSet<BasicBlock*, 16> >
- I = idf_ext_begin(*PI, TranspBlocks),
- E = idf_ext_end(*PI, TranspBlocks); I != E; ++I)
- if (AA.canBasicBlockModify(**I, Arg, LoadSize))
+ I = idf_ext_begin(P, TranspBlocks),
+ E = idf_ext_end(P, TranspBlocks); I != E; ++I)
+ if (AA.canBasicBlockModify(**I, Loc))
return false;
+ }
}
// If the path from the entry of the function to each load is free of
/// DoPromotion - This method actually performs the promotion of the specified
/// arguments, and returns the new function. At this point, we know that it's
/// safe to do so.
-Function *ArgPromotion::DoPromotion(Function *F,
-
SmallPtrSet<Argument*, 8> &ArgsToPromote,
+CallGraphNode *ArgPromotion::DoPromotion(Function *F,
+ SmallPtrSet<Argument*, 8> &ArgsToPromote,
SmallPtrSet<Argument*, 8> &ByValArgsToTransform) {
// Start by computing a new prototype for the function, which is the same as
// the old function, but has modified arguments.
- const FunctionType *FTy = F->getFunctionType();
- std::vector<const Type*> Params;
+ FunctionType *FTy = F->getFunctionType();
+ std::vector<Type*> Params;
typedef std::set<IndicesVector> ScalarizeTable;
// what the new GEP/Load instructions we are inserting look like.
std::map<IndicesVector, LoadInst*> OriginalLoads;
- // ParamAttrs - Keep track of the parameter attributes for the arguments
+ // Attribute - Keep track of the parameter attributes for the arguments
// that we are *not* promoting. For the ones that we do promote, the parameter
// attributes are lost
- SmallVector<ParamAttrsWithIndex, 8> ParamAttrsVec;
- const PAListPtr &PAL = F->getParamAttrs();
+ SmallVector<AttributeWithIndex, 8> AttributesVec;
+ const AttributeSet &PAL = F->getAttributes();
// Add any return attributes.
- if (ParameterAttributes attrs = PAL.getParamAttrs(0))
- ParamAttrsVec.push_back(ParamAttrsWithIndex::get(0, attrs));
-
+ Attribute attrs = PAL.getRetAttributes();
+ if (attrs.hasAttributes())
+ AttributesVec.push_back(AttributeWithIndex::get(AttributeSet::ReturnIndex,
+ attrs));
+
// First, determine the new argument list
unsigned ArgIndex = 1;
for (Function::arg_iterator I = F->arg_begin(), E = F->arg_end(); I != E;
++I, ++ArgIndex) {
if (ByValArgsToTransform.count(I)) {
// Simple byval argument? Just add all the struct element types.
-
const Type *AgTy = cast<PointerType>(I->getType())->getElementType();
- const StructType *STy = cast<StructType>(AgTy);
+ Type *AgTy = cast<PointerType>(I->getType())->getElementType();
+ StructType *STy = cast<StructType>(AgTy);
for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i)
Params.push_back(STy->getElementType(i));
++NumByValArgsPromoted;
} else if (!ArgsToPromote.count(I)) {
// Unchanged argument
Params.push_back(I->getType());
- if (ParameterAttributes attrs = PAL.getParamAttrs(ArgIndex))
- ParamAttrsVec.push_back(ParamAttrsWithIndex::get(Params.size(), attrs));
+ Attribute attrs = PAL.getParamAttributes(ArgIndex);
+ if (attrs.hasAttributes())
+ AttributesVec.push_back(AttributeWithIndex::get(Params.size(), attrs));
} else if (I->use_empty()) {
// Dead argument (which are always marked as promotable)
++NumArgumentsDead;
// Add a parameter to the function for each element passed in.
for (ScalarizeTable::iterator SI = ArgIndices.begin(),
E = ArgIndices.end(); SI != E; ++SI) {
- Params.push_back(GetElementPtrInst::getIndexedType(I->getType(),
- &*SI->begin(),
- SI->size()));
+ // not allowed to dereference ->begin() if size() is 0
+ Params.push_back(GetElementPtrInst::getIndexedType(I->getType(), *SI));
assert(Params.back());
}
}
}
- const Type *RetTy = FTy->getReturnType();
+ // Add any function attributes.
+ attrs = PAL.getFnAttributes();
+ if (attrs.hasAttributes())
+ AttributesVec.push_back(AttributeWithIndex::get(AttributeSet::FunctionIndex,
+ attrs));
- // Work around LLVM bug PR56: the CWriter cannot emit varargs functions which
- // have zero fixed arguments.
- bool ExtraArgHack = false;
- if (Params.empty() && FTy->isVarArg()) {
- ExtraArgHack = true;
- Params.push_back(Type::Int32Ty);
- }
+ Type *RetTy = FTy->getReturnType();
// Construct the new function type using the new arguments.
FunctionType *NFTy = FunctionType::get(RetTy, Params, FTy->isVarArg());
- // Create the new function body and insert it into the module...
+ // Create the new function body and insert it into the module.
Function *NF = Function::Create(NFTy, F->getLinkage(), F->getName());
NF->copyAttributesFrom(F);
+
+ DEBUG(dbgs() << "ARG PROMOTION: Promoting to:" << *NF << "\n"
+ << "From: " << *F);
+
// Recompute the parameter attributes list based on the new arguments for
// the function.
- NF->setParamAttrs(PAListPtr::get(ParamAttrsVec.begin(), ParamAttrsVec.end()));
- ParamAttrsVec.clear();
+ NF->setAttributes(AttributeSet::get(F->getContext(), AttributesVec));
+ AttributesVec.clear();
F->getParent()->getFunctionList().insert(F, NF);
NF->takeName(F);
// changes.
AliasAnalysis &AA = getAnalysis<AliasAnalysis>();
+ // Get the callgraph information that we need to update to reflect our
+ // changes.
+ CallGraph &CG = getAnalysis<CallGraph>();
+
+ // Get a new callgraph node for NF.
+ CallGraphNode *NF_CGN = CG.getOrInsertFunction(NF);
+
// Loop over all of the callers of the function, transforming the call sites
// to pass in the loaded pointers.
//
SmallVector<Value*, 16> Args;
while (!F->use_empty()) {
- CallSite CS = CallSite::get(F->use_back());
+ CallSite CS(F->use_back());
+ assert(CS.getCalledFunction() == F);
Instruction *Call = CS.getInstruction();
- const PAListPtr &CallPAL = CS.getParamAttrs();
-
+ const AttributeSet &CallPAL = CS.getAttributes();
+
// Add any return attributes.
- if (ParameterAttributes attrs = CallPAL.getParamAttrs(0))
- ParamAttrsVec.push_back(ParamAttrsWithIndex::get(0, attrs));
+ Attribute attrs = CallPAL.getRetAttributes();
+ if (attrs.hasAttributes())
+ AttributesVec.push_back(AttributeWithIndex::get(AttributeSet::ReturnIndex,
+ attrs));
// Loop over the operands, inserting GEP and loads in the caller as
// appropriate.
I != E; ++I, ++AI, ++ArgIndex)
if (!ArgsToPromote.count(I) && !ByValArgsToTransform.count(I)) {
Args.push_back(*AI); // Unmodified argument
-
- if (ParameterAttributes Attrs = CallPAL.getParamAttrs(ArgIndex))
- ParamAttrsVec.push_back(ParamAttrsWithIndex::get(Args.size(), Attrs));
-
+
+ Attribute Attrs = CallPAL.getParamAttributes(ArgIndex);
+ if (Attrs.hasAttributes())
+ AttributesVec.push_back(AttributeWithIndex::get(Args.size(), Attrs));
+
} else if (ByValArgsToTransform.count(I)) {
// Emit a GEP and load for each element of the struct.
- const Type *AgTy = cast<PointerType>(I->getType())->getElementType();
- const StructType *STy = cast<StructType>(AgTy);
- Value *Idxs[2] = { ConstantInt::get(Type::Int32Ty, 0), 0 };
+ Type *AgTy = cast<PointerType>(I->getType())->getElementType();
+ StructType *STy = cast<StructType>(AgTy);
+ Value *Idxs[2] = {
+ ConstantInt::get(Type::getInt32Ty(F->getContext()), 0), 0 };
for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
- Idxs[1] = ConstantInt::get(Type::Int32Ty, i);
- Value *Idx = GetElementPtrInst::Create(*AI, Idxs, Idxs+2,
+ Idxs[1] = ConstantInt::get(Type::getInt32Ty(F->getContext()), i);
+ Value *Idx = GetElementPtrInst::Create(*AI, Idxs,
(*AI)->getName()+"."+utostr(i),
Call);
// TODO: Tell AA about the new values?
Args.push_back(new LoadInst(Idx, Idx->getName()+".val", Call));
- }
+ }
} else if (!I->use_empty()) {
// Non-dead argument: insert GEPs and loads as appropriate.
ScalarizeTable &ArgIndices = ScalarizedElements[I];
// Store the Value* version of the indices in here, but declare it now
- // for reuse
+ // for reuse.
std::vector<Value*> Ops;
for (ScalarizeTable::iterator SI = ArgIndices.begin(),
E = ArgIndices.end(); SI != E; ++SI) {
LoadInst *OrigLoad = OriginalLoads[*SI];
if (!SI->empty()) {
Ops.reserve(SI->size());
- const Type *ElTy = V->getType();
- for (IndicesVector::const_iterator II = SI->begin(),
+ Type *ElTy = V->getType();
+ for (IndicesVector::const_iterator II = SI->begin(),
IE = SI->end(); II != IE; ++II) {
// Use i32 to index structs, and i64 for others (pointers/arrays).
// This satisfies GEP constraints.
- const Type *IdxTy = (isa<StructType>(ElTy) ? Type::Int32Ty : Type::Int64Ty);
+ Type *IdxTy = (ElTy->isStructTy() ?
+ Type::getInt32Ty(F->getContext()) :
+ Type::getInt64Ty(F->getContext()));
Ops.push_back(ConstantInt::get(IdxTy, *II));
- // Keep track of the type we're currently indexing
+ // Keep track of the type we're currently indexing.
ElTy = cast<CompositeType>(ElTy)->getTypeAtIndex(*II);
}
- // And create a GEP to extract those indices
- V = GetElementPtrInst::Create(V, Ops.begin(), Ops.end(),
- V->getName()+".idx", Call);
+ // And create a GEP to extract those indices.
+ V = GetElementPtrInst::Create(V, Ops, V->getName()+".idx", Call);
Ops.clear();
AA.copyValue(OrigLoad->getOperand(0), V);
}
- Args.push_back(new LoadInst(V, V->getName()+".val", Call));
+ // Since we're replacing a load make sure we take the alignment
+ // of the previous load.
+ LoadInst *newLoad = new LoadInst(V, V->getName()+".val", Call);
+ newLoad->setAlignment(OrigLoad->getAlignment());
+ // Transfer the TBAA info too.
+ newLoad->setMetadata(LLVMContext::MD_tbaa,
+ OrigLoad->getMetadata(LLVMContext::MD_tbaa));
+ Args.push_back(newLoad);
AA.copyValue(OrigLoad, Args.back());
}
}
- if (ExtraArgHack)
- Args.push_back(Constant::getNullValue(Type::Int32Ty));
-
- // Push any varargs arguments on the list
+ // Push any varargs arguments on the list.
for (; AI != CS.arg_end(); ++AI, ++ArgIndex) {
Args.push_back(*AI);
- if (ParameterAttributes Attrs = CallPAL.getParamAttrs(ArgIndex))
- ParamAttrsVec.push_back(ParamAttrsWithIndex::get(Args.size(), Attrs));
+ Attribute Attrs = CallPAL.getParamAttributes(ArgIndex);
+ if (Attrs.hasAttributes())
+ AttributesVec.push_back(AttributeWithIndex::get(Args.size(), Attrs));
}
+ // Add any function attributes.
+ attrs = CallPAL.getFnAttributes();
+ if (attrs.hasAttributes())
+ AttributesVec.push_back(AttributeWithIndex::get(AttributeSet::FunctionIndex,
+ attrs));
+
Instruction *New;
if (InvokeInst *II = dyn_cast<InvokeInst>(Call)) {
New = InvokeInst::Create(NF, II->getNormalDest(), II->getUnwindDest(),
- Args.begin(), Args.end(), "", Call);
+ Args, "", Call);
cast<InvokeInst>(New)->setCallingConv(CS.getCallingConv());
- cast<InvokeInst>(New)->setParamAttrs(PAListPtr::get(ParamAttrsVec.begin(),
- ParamAttrsVec.end()));
+ cast<InvokeInst>(New)->setAttributes(AttributeSet::get(II->getContext(),
+ AttributesVec));
} else {
- New = CallInst::Create(NF, Args.begin(), Args.end(), "", Call);
+ New = CallInst::Create(NF, Args, "", Call);
cast<CallInst>(New)->setCallingConv(CS.getCallingConv());
- cast<CallInst>(New)->setParamAttrs(PAListPtr::get(ParamAttrsVec.begin(),
- ParamAttrsVec.end()));
+ cast<CallInst>(New)->setAttributes(AttributeSet::get(New->getContext(),
+ AttributesVec));
if (cast<CallInst>(Call)->isTailCall())
cast<CallInst>(New)->setTailCall();
}
Args.clear();
- ParamAttrsVec.clear();
+ AttributesVec.clear();
// Update the alias analysis implementation to know that we are replacing
// the old call with a new one.
AA.replaceWithNewValue(Call, New);
+ // Update the callgraph to know that the callsite has been transformed.
+ CallGraphNode *CalleeNode = CG[Call->getParent()->getParent()];
+ CalleeNode->replaceCallEdge(Call, New, NF_CGN);
+
if (!Call->use_empty()) {
Call->replaceAllUsesWith(New);
New->takeName(Call);
// function empty.
NF->getBasicBlockList().splice(NF->begin(), F->getBasicBlockList());
- // Loop over the argument list, transfering uses of the old arguments over to
- // the new arguments, also transfering over the names as well.
+ // Loop over the argument list, transferring uses of the old arguments over to
+ // the new arguments, also transferring over the names as well.
//
for (Function::arg_iterator I = F->arg_begin(), E = F->arg_end(),
I2 = NF->arg_begin(); I != E; ++I) {
++I2;
continue;
}
-
+
if (ByValArgsToTransform.count(I)) {
// In the callee, we create an alloca, and store each of the new incoming
// arguments into the alloca.
Instruction *InsertPt = NF->begin()->begin();
-
+
// Just add all the struct element types.
-
const Type *AgTy = cast<PointerType>(I->getType())->getElementType();
+ Type *AgTy = cast<PointerType>(I->getType())->getElementType();
Value *TheAlloca = new AllocaInst(AgTy, 0, "", InsertPt);
- const StructType *STy = cast<StructType>(AgTy);
- Value *Idxs[2] = { ConstantInt::get(Type::Int32Ty, 0), 0 };
-
+ StructType *STy = cast<StructType>(AgTy);
+ Value *Idxs[2] = {
+ ConstantInt::get(Type::getInt32Ty(F->getContext()), 0), 0 };
+
for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
- Idxs[1] = ConstantInt::get(Type::Int32Ty, i);
- std::string Name = TheAlloca->getName()+"."+utostr(i);
- Value *Idx = GetElementPtrInst::Create(TheAlloca, Idxs, Idxs+2,
- Name, InsertPt);
- I2->setName(I->getName()+"."+utostr(i));
+ Idxs[1] = ConstantInt::get(Type::getInt32Ty(F->getContext()), i);
+ Value *Idx =
+ GetElementPtrInst::Create(TheAlloca, Idxs,
+ TheAlloca->getName()+"."+Twine(i),
+ InsertPt);
+ I2->setName(I->getName()+"."+Twine(i));
new StoreInst(I2++, Idx, InsertPt);
}
-
+
// Anything that used the arg should now use the alloca.
I->replaceAllUsesWith(TheAlloca);
TheAlloca->takeName(I);
AA.replaceWithNewValue(I, TheAlloca);
continue;
- }
-
+ }
+
if (I->use_empty()) {
AA.deleteValue(I);
continue;
}
-
+
// Otherwise, if we promoted this argument, then all users are load
// instructions (or GEPs with only load users), and all loads should be
// using the new argument that we added.
LI->replaceAllUsesWith(I2);
AA.replaceWithNewValue(LI, I2);
LI->eraseFromParent();
- DOUT << "*** Promoted load of argument '" << I->getName()
- << "' in function '" << F->getName() << "'\n";
+ DEBUG(dbgs() << "*** Promoted load of argument '" << I->getName()
+ << "' in function '" << F->getName() << "'\n");
} else {
GetElementPtrInst *GEP = cast<GetElementPtrInst>(I->use_back());
IndicesVector Operands;
NewName += ".val";
TheArg->setName(NewName);
- DOUT << "*** Promoted agg argument '" << TheArg->getName()
- << "' of function '" << NF->getName() << "'\n";
+ DEBUG(dbgs() << "*** Promoted agg argument '" << TheArg->getName()
+ << "' of function '" << NF->getName() << "'\n");
// All of the uses must be load instructions. Replace them all with
// the argument specified by ArgNo.
}
// Increment I2 past all of the arguments added for this promoted pointer.
- for (unsigned i = 0, e = ArgIndices.size(); i != e; ++i)
- ++I2;
+ std::advance(I2, ArgIndices.size());
}
- // Notify the alias analysis implementation that we inserted a new argument.
- if (ExtraArgHack)
- AA.copyValue(Constant::getNullValue(Type::Int32Ty), NF->arg_begin());
-
-
// Tell the alias analysis that the old function is about to disappear.
AA.replaceWithNewValue(F, NF);
- // Now that the old function is dead, delete it.
- F->eraseFromParent();
- return NF;
+
+ NF_CGN->stealCalledFunctionsFrom(CG[F]);
+
+ // Now that the old function is dead, delete it. If there is a dangling
+ // reference to the CallgraphNode, just leave the dead function around for
+ // someone else to nuke.
+ CallGraphNode *CGN = CG[F];
+ if (CGN->getNumReferences() == 0)
+ delete CG.removeFunctionFromModule(CGN);
+ else
+ F->setLinkage(Function::ExternalLinkage);
+
+ return NF_CGN;
}
projects / oota-llvm.git / blobdiff